Posts

• ---

  Local LMs
  M montez

  Qwen-Image-2.1

  Hugging Face: https://huggingface.co/Qwen/Qwen-Image-2.1, https://huggingface.co/Qwen/Qwen-Image-2.1-Turbo
  GitHub: https://github.com/QwenLM/Qwen-Image-2.1
  Website: https://qwen.ai/blog?id=qwen-image-2.1

  Developer: Alibaba Cloud, Qwen team
  Released: September 2026
  Variants: Qwen-Image-2.1 (40 default denoising steps) / Qwen-Image-2.1-Turbo (8 denoising steps)
  Parameters: 7B visual generation component, plus Qwen3-VL 8B text encoder
  Resolution: native 2K, 2048x2048 default; recommended sizes include 2752x1536 and 1536x2752
  Architecture: single-stream DiT, 32 layers, block-causal attention with prefix KV cache reuse; Qwen3-VL 8B text encoder; 64-channel RGBA VAE with 16x spatial compression; flow matching with Euler scheduler
  License: Qwen Research License Agreement
  Modalities: Text-to-Image + Image editing (up to 10 reference images), native RGBA transparency
  Runs on: Desktop GPU, 24GB+ VRAM with model CPU offload
  Formats: BF16 safetensors, Diffusers pipeline
  On disk: Qwen-Image-2.1: 33.13GB (14.23GB transformer, 17.53GB text encoder, 1.35GB VAE) / Turbo: 32.49GB (14.23GB transformer, 17.53GB text encoder, 0.68GB VAE)

  AI & Software llm foundation models

• ---

  Local LMs
  M montez

  NII LLM-jp-4

  Hugging Face: https://huggingface.co/llm-jp
  GitHub: https://github.com/llm-jp/llm-jp-4-cookbook
  Website: https://llm-jp.nii.ac.jp/blog/llm-jp-4-1/

  Developer: National Institute of Informatics, Research and Development Center for Large Language Models (LLMC), LLM-jp
  Released: April-September 2026, rolling family
  Variants: LLM-jp-4 8B and 32B-A3B (April 2026) / LLM-jp-4 33B (August 2026) / LLM-jp-4-VL 9B (September 2026) / LLM-jp-4.1 8B, 32B-A3B, 33B Thinking (September 2026)
  Parameters: 8B: 8.59B / 32B-A3B: 32.14B total, 3.83B active / 33B: 33.22B / VL 9B: 8.6B language model + 0.4B vision encoder
  Context: 65,536 tokens, text models
  Architecture: 8B and 33B: dense Llama architecture, 32 and 64 layers / 32B-A3B: Qwen3-MoE architecture, 32 layers, 128 routed experts with 8 active / VL 9B: llm-jp-4-8b-thinking, SigLIP 2 So400m vision encoder, 2-layer MLP projector
  License: Apache 2.0
  Modalities: 8B, 32B-A3B, 33B: Text / VL 9B: Text + Image
  Runs on: 8B: Laptop / 32B-A3B and 33B: Laptop, 32GB+ unified memory / VL 9B: Laptop, 24GB+ unified memory
  Formats: safetensors BF16, GGUF (BF16, Q4_K_M; LLM-jp-4.1 text models)
  On disk: LLM-jp-4.1 8B: 5.50GB Q4_K_M / 32B-A3B: 21.52GB Q4_K_M / 33B: 20.41GB Q4_K_M / VL 9B: 18.11GB BF16

  AI & Software llm foundation models

• ---

  Embodied Foundation Models
  M montez

  Metacognition NavGPT3

  Hugging Face: https://huggingface.co/Metacognition-AI/NavGPT3-8B, https://huggingface.co/Metacognition-AI/NavGPT3-4B
  GitHub: https://github.com/metacognitionai/NavGPT-3
  Website: https://metacognitionai.github.io/NavGPT3/
  Technical report: https://arxiv.org/abs/2610.10787

  Developer: Metacognition
  Released: October 2026
  Variants: 4B, 8B
  Parameters: 4.44B (4B checkpoint), 8.77B (8B checkpoint)
  Architecture: Qwen3-VL fine-tune with a two-layer MLP action head on the last prompt token’s hidden state; one 3,072-token visual budget shared across a four-view, up to 16-step image history
  License: GNU Affero General Public License v3.0
  Modalities: Text + Four-view RGB image history in / 8 waypoints (x, y, theta) out
  Runs on: NVIDIA GPU (Linux)
  Formats: safetensors
  On disk: 9.66GB (4B checkpoint), 17.54GB (8B checkpoint)

  AI & Software embodied foundation models robotics

• ---

  Local LMs
  M montez

  OpenBMB MiniCPM5

  Hugging Face: https://huggingface.co/openbmb/MiniCPM5-2B, https://huggingface.co/openbmb/MiniCPM5-1B
  GitHub: https://github.com/OpenBMB/MiniCPM

  Developer: OpenBMB
  Released: May-September 2026, rolling family
  Variants: MiniCPM5-2B (September 2026) / MiniCPM5-1B (May 2026)
  Parameters: 2B: 2.52B, 1.98B non-embedding / 1B: 1.08B, 0.68B non-embedding
  Context: 131,072 tokens
  Architecture: dense LlamaForCausalLM, GQA with 16 Q and 2 KV heads; 2B: 42 layers / 1B: 24 layers
  License: Apache 2.0
  Modalities: Text
  Runs on: Smartphone, Laptop
  Formats: safetensors BF16, GGUF (F16, Q8_0, Q4_K_M), MLX 4-bit, GPTQ 4-bit (2B)
  On disk: 2B: 1.56GB Q4_K_M, 5.03GB BF16 / 1B: 0.69GB Q4_K_M, 2.16GB BF16

  AI & Software llm foundation models

• ---

  Embodied Foundation Models
  M montez

  DeepCybo PhysBrain 1.5

  Hugging Face: https://huggingface.co/DeepCybo/PhysBrain1.5-8B, https://huggingface.co/DeepCybo/PhysBrain1.5-2B
  GitHub: https://github.com/DeepCybo-PhysAI/PhysBrain-1.5
  Website: https://deepcybo-physai.github.io/PhysBrain-1.5/
  Technical report: https://arxiv.org/abs/2609.14973

  Developer: DeepCybo, Zhongguancun Academy, Zhongguancun Institute of Artificial Intelligence
  Released: September 2026
  Variants: 2B, 8B
  Parameters: 2.16B (2B checkpoint), 8.90B (8B checkpoint)
  Context: 262,144 tokens
  Architecture: Qwen3-VL backbone extended with action and visual-state tokens; one shared autoregressive transformer, single next-token objective, no task-specific heads; ActionPiece action tokens with one action codebook across robot setups
  Modalities: Text + Image + Video + Action history in / Text + Spatial grounding + Action chunk + Future-state image, depth map, robot mask out
  Runs on: 2B: Laptop, Desktop GPU / 8B: Desktop GPU
  Formats: safetensors
  On disk: 4.32GB (2B checkpoint), 17.81GB (8B checkpoint)

  AI & Software embodied foundation models robotics

• ---

  Local LMs
  M montez

  SparkLLM Spark-X2.5

  Hugging Face: https://huggingface.co/XHToken/Spark-X2.5-4B, https://huggingface.co/XHToken/Spark-X2.5-1.7B
  GitHub: https://github.com/XHToken/Spark-X2.5
  Website: https://dev.to/sparkllm/spark-x25-4b-17b-the-only-on-device-models-with-native-1m-token-context-now-open-source-d9o

  Developer: SparkLLM
  Released: September 2026
  Variants: 4B / 1.7B
  Parameters: 4B / 1.7B
  Context: 1,048,576 tokens
  Architecture: dense; hybrid attention with 1 full-attention layer to 3 sliding-window layers, window 512; 4B: 36 layers, GQA 16 Q and 4 KV heads / 1.7B: 28 layers, GQA 8 Q and 2 KV heads
  License: Apache 2.0
  Modalities: Text
  Runs on: Smartphone, Laptop, Edge device
  Formats: safetensors BF16, FP8, INT8, GGUF (Q4_K_M, Q8_0, F16)
  On disk: 4B: 2.60GB Q4_K_M / 1.7B: 1.11GB Q4_K_M

  AI & Software llm foundation models

• ---

  Embodied Foundation Models
  M montez

  Meta FAIR RoboJEPA

  GitHub: https://github.com/facebookresearch/robo_jepa
  Website: https://robojepa.github.io
  Technical report: https://arxiv.org/abs/2610.10515

  Developer: Meta FAIR
  Released: October 2026
  Variants: 22M / 50M / 100M / 300M / 1B / 2B / 4B / 8B / 8B DROID 720p (3 views)
  Parameters: 22M to 8B predictors on a frozen V-JEPA 2.1 ViT-G/384 encoder
  Architecture: Action-conditioned JEPA latent world model: a transformer predictor over frozen V-JEPA 2.1 features, with latent CEM/MPC planning toward a goal image and an optional diffusion video decoder
  License: CC BY-NC-SA 4.0
  Modalities: Camera video + action sequence in / predicted future latent states out (optional decoded video); no action head, actions come from latent MPC planning
  Runs on: NVIDIA GPU (CUDA 12.6), one GPU per model size for evaluation; no VRAM requirement published
  Formats: PyTorch (.pth.tar) from dl.fbaipublicfiles.com
  On disk: 8B 29.99GB, 1B 4.07GB, 300M 1.24GB (300k-step checkpoints), plus V-JEPA 2.1 ViT-G/384 encoder 30.24GB

  AI & Software embodied foundation models robotics

• ---

  Local LMs
  M montez

  China Telecom Xing4.0

  Hugging Face: https://huggingface.co/XingChen-AGI/Xing4.0-29B-A4B, https://huggingface.co/XingChen-AGI/Xing4.0-29B-A4B-GGUF, https://huggingface.co/XingChen-AGI/Xing4.0-29B-A4B-FP8
  GitHub: https://github.com/XingChen-AGI/Xing4.0-29B-A4B

  Developer: China Telecom Artificial Intelligence Technology Co., Ltd.
  Released: September 2026
  Parameters: 29B total, 4B active
  Context: 262,144 tokens, extensible to 512K
  Architecture: MoE with mHC, MLA attention and MTP; 40 layers, 64 routed experts with 4 active plus 1 shared; trained on Ascend NPUs with MindSpore
  License: Apache 2.0
  Modalities: Text
  Runs on: Desktop GPU
  Formats: safetensors BF16, FP8, GGUF IQ4_NL
  On disk: 20.1GB IQ4_NL / 33.17GB FP8 / 62.43GB BF16

  AI & Software llm foundation models

• ---

  Embodied Foundation Models
  M montez

  Microsoft Rho

  Hugging Face: https://huggingface.co/microsoft/rho-base, https://huggingface.co/collections/microsoft/rho
  GitHub: https://github.com/microsoft/rhobotics
  Website: https://microsoft.github.io/rhobotics/
  Technical report: https://arxiv.org/abs/2609.38164

  Developer: Microsoft Research
  Released: September 2026
  Variants: rho-base / rho-yam-box / rho-ur-ai-trainer / rho-fr3-duo / rho-libero / rho-roboeval
  Parameters: 5B (Phi-Phy VLM 4.68B / flow-matching action expert 542M)
  Architecture: Vision-language-action model: a physically grounded Phi-family VLM (Phi-Phy) with a 12-block flow-matching action expert that cross-attends to a VLM decoder layer, midtrained per embodiment
  License: MIT
  Modalities: Multi-camera images + language instruction + robot state in / action chunk out (up to 50 steps in pretraining)
  Runs on: NVIDIA GPU (Linux, CUDA; FlashAttention 2 by default); no VRAM requirement published
  Formats: safetensors (BF16, 3 shards)
  On disk: rho-base 10.50GB

  AI & Software embodied foundation models robotics

• ---

  Local LMs
  M montez

  AI Singapore Nemotron-SEA-LION v4.8 30B-A3B

  Hugging Face: https://huggingface.co/aisingapore/Nemotron-SEA-LION-v4.8-30B-A3B, https://huggingface.co/aisingapore/Nemotron-SEA-LION-v4.8-30B-A3B-GGUF, https://huggingface.co/aisingapore/Nemotron-SEA-LION-v4.8-30B-A3B-FP8, https://huggingface.co/aisingapore/Nemotron-SEA-LION-v4.8-30B-A3B-NVFP4
  Website: https://sea-lion.ai/blog/uplifting-ai-in-southeast-asia-sea-announcing-nemotron-sea-lion-v4-8-in-collaboration-with-nvidia/
  Technical report: https://arxiv.org/abs/2609.18310

  Developer: AI Singapore, with NVIDIA
  Released: September 2026
  Parameters: 30B total, 3B active
  Context: 262,144 tokens
  Architecture: Mamba2-Transformer hybrid MoE; NVIDIA Nemotron 3 Nano 30B-A3B base, continued pretraining on 150B tokens, then SFT and on-policy distillation; English plus 7 Southeast Asian languages
  License: MIT
  Modalities: Text
  Runs on: Laptop, 32GB+ unified memory
  Formats: safetensors BF16, FP8, NVFP4, GGUF (Q4_K_M, Q6_K, Q8_0, F16)
  On disk: 25.43GB Q4_K_M / 22.94GB NVFP4 / 34.96GB FP8 / 65.83GB BF16

  AI & Software llm foundation models